The Agent Skills Directory

[PROMPT_INJECTION]: The skill identifies and processes untrusted project data and local templates, creating a potential surface for indirect prompt injection.
Ingestion points: The skill uses Read and Grep to analyze project files and loads templates from the .kiro/settings/templates/steering-custom/ directory.
Boundary markers: The instructions do not define specific delimiters or isolation techniques (e.g., XML tags or clear boundary warnings) for the data ingested from the project codebase.
Capability inventory: The agent has access to file reading and searching tools (Read, Glob, Grep), file writing capabilities (creating files in .kiro/steering/), and basic shell command execution (Bash with ls).
Sanitization: While the steering-principles.md file explicitly instructs the agent to avoid secrets and credentials, there are no instructions to sanitize or escape the technical content of files to prevent the agent from accidentally following instructions embedded within the codebase or templates.

kiro-steering-custom